[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"branding":3,"analytics":7,"article-new-world-model-blends-memory-and-physics-for-planning":10,"sections":35},{"siteName":4,"siteTagline":5,"publisherName":4,"contactEmail":6},"The Revision","Tech news, decoded.","editor@therevision.news",{"gaMeasurementId":8,"adsenseClientId":9},"G-ZW2MV82GYR","ca-pub-8533917693782264",{"article":11},{"id":12,"slug":13,"title":14,"dek":15,"body_md":16,"tags_json":17,"published_at":18,"created_at":19,"updated_at":20,"status":21,"review_note":22,"review_notes":23,"image_url":22,"persona_id":22,"persona_name":22,"section":24,"tags":25,"sources":30,"feedback":34,"feedback_at":22,"cost_usd":34,"total_tokens":34},8101,"new-world-model-blends-memory-and-physics-for-planning","New World Model Blends Memory and Physics for Planning","A new world model combines selective memory and Hamiltonian physics to cut short-term planning errors, though long-horizon stability remains unresolved.","A new world model for AI planning splits the difference between physics and memory - and mostly wins.\n\nResearchers built HaM-World, a world model that pairs Mamba-based selective memory with a Soft-Hamiltonian latent dynamics prior. The system splits its internal state into a canonical physics-like subspace and a separate context subspace for semantic and non-conservative factors. Tested across six DeepMind Control Suite tasks, it ranked first on four and second on two, posting the highest average AUC on the core four-task suite - 9.5% above the TD-MPC2 baseline. For the short-to-medium planning horizons the system actually uses, it cut imagined-rollout error to 45% of that baseline and won 11 of 12 comparisons at horizons of 3, 5, and 7 steps; at longer horizons past step 7, its error grows faster and other methods catch up.\n\nThat honesty matters. World models are supposed to let an agent imagine outcomes before acting, but those imagined rollouts often drift into nonsense as the horizon stretches. HaM-World does not solve that - it just pushes the failure point further out and holds up better under stress along the way. Under 12 out-of-distribution tests involving action delay, masked observations, and shifted dynamics, it posted the best returns in every single condition, including a 13.6% gain on the Reacher Easy task.\n\nAblations credit most of the improvement to the memory component, with the Hamiltonian structure adding smaller, consistent gains on top - a reminder that the fancier-sounding idea is not always doing the heavy lifting.","[\"world-models\",\"reinforcement-learning\",\"ai-research\",\"planning\"]","2026-09-28T04:00:00.000Z","2026-09-28T09:32:15.021Z","2026-09-28T09:32:22.962Z","published",null,[],"ai",[26,27,28,29],"world-models","reinforcement-learning","ai-research","planning",[31],{"name":32,"url":33},"arXiv cs.AI","https:\u002F\u002Farxiv.org\u002Fabs\u002F2605.05951",0,{"sections":36},[37,40,44,49,54,59,63,68,73,78,83,88,92,97],{"name":38,"slug":24,"count":39,"latest_published_at":18},"AI",4791,{"name":41,"slug":42,"count":43,"latest_published_at":18},"Security","security",762,{"name":45,"slug":46,"count":47,"latest_published_at":48},"Policy","policy",399,"2026-09-27T18:39:02.000Z",{"name":50,"slug":51,"count":52,"latest_published_at":53},"Deals","deals",261,"2026-09-27T15:30:35.000Z",{"name":55,"slug":56,"count":57,"latest_published_at":58},"Hardware","hardware",188,"2026-09-27T20:46:36.000Z",{"name":60,"slug":61,"count":62,"latest_published_at":18},"Science","science",151,{"name":64,"slug":65,"count":66,"latest_published_at":67},"Consumer Tech","consumer-tech",135,"2026-09-26T14:30:00.000Z",{"name":69,"slug":70,"count":71,"latest_published_at":72},"Software","software",91,"2026-09-25T20:55:00.000Z",{"name":74,"slug":75,"count":76,"latest_published_at":77},"Dev Tools","dev-tools",84,"2026-09-26T04:20:58.000Z",{"name":79,"slug":80,"count":81,"latest_published_at":82},"Startups","startups",76,"2026-09-25T18:33:59.000Z",{"name":84,"slug":85,"count":86,"latest_published_at":87},"Gaming","gaming",48,"2026-09-25T18:35:21.000Z",{"name":89,"slug":90,"count":86,"latest_published_at":91},"General","general","2026-09-26T17:02:42.000Z",{"name":93,"slug":94,"count":95,"latest_published_at":96},"Reviews","reviews",30,"2026-09-24T20:07:31.000Z",{"name":98,"slug":99,"count":100,"latest_published_at":101},"How-To","how-to",6,"2026-06-16T09:00:00.000Z"]